AI optical stock Lumentum soars over 120 percent in 2026

While Nvidia (NASDAQ: NVDA) remains a dominant force in artificial intelligence, its stock performance is being dwarfed by optical components maker Lumentum Holdings (NASDAQ: LITE), which has seen its shares gain over 120 percent year-to-date in 2026. The surge indicates a broadening of the AI investment thesis beyond GPU designers to the critical "picks and shovels" suppliers building the network infrastructure for large-scale AI.

The bullish sentiment extends across the optical supply chain. Rosenblatt recently raised its price target on manufacturing partner Fabrinet (NYSE: FN) to $715 from $550, citing increased investment by AI leaders as a positive signal for the broader ecosystem. The firm noted that Fabrinet is a critical manufacturing partner for the advanced transceivers and co-packaged optics modules needed to meet accelerating demand for AI networking.

The initial AI boom was driven by the need for raw computing power to train large language models, a race that overwhelmingly benefited Nvidia. However, the next phase of AI focuses on networking massive clusters of chips, inference, and agentic AI. With AI clusters now scaling beyond one million individual chips, the performance of the network is as critical as the performance of a single processor. This shift puts a premium on components that reduce congestion and slash energy consumption.

This second wave directly benefits companies like Broadcom (NASDAQ: AVGO), a leader in the Ethernet switches and optical interconnects that form the backbone of data center networking. It also buoys Arista Networks (NYSE: ANET), which packages these components with its Extensible Operating System (EOS) to provide manageable, high-performance networking platforms for hyperscalers like Microsoft (NASDAQ: MSFT) and Meta Platforms (NASDAQ: META).

The Rise of the AI Network

As AI models grow, the challenge shifts from pure computation to efficient data movement. Broadcom's leadership in high-bandwidth Ethernet switching solutions is central to this new phase. The company's components are integral to preventing bottlenecks in AI systems that can cripple performance, regardless of the power of the underlying chips. Arista Networks, a key customer of Broadcom, integrates this hardware into comprehensive solutions for its top clients, demonstrating the deep supply chain integration at play.

The trend also includes a move toward custom-designed chips. Alphabet's (NASDAQ: GOOGL) partnership with Broadcom to develop its Tensor Processing Units (TPUs) is a prime example. These purpose-built AI ASICs give Google a structural cost advantage, with SemiAnalysis estimating a 44 percent lower total cost of ownership compared to equivalent Nvidia-based servers. This efficiency is crucial for high-volume inference, the process of running trained AI models.

Custom Silicon and Manufacturing Scale

The demand for custom, high-performance optical solutions is pushing manufacturing capabilities to new limits. Fabrinet, a key partner for many component designers, is directly addressing this by establishing a dedicated manufacturing line for silicon photonics-based optical circuit switching systems. This expansion, expected to be fully operational in Q2 2026, is designed to support the growing demand from hyperscalers and AI system providers.

Looking forward, the rise of agentic AI—where AI agents perform tasks independently—is expected to create another demand shock for data center infrastructure. This trend will require a massive increase in high-performance CPUs that provide sequential logic and reasoning. Both Advanced Micro Devices (NASDAQ: AMD), the current leader in data center CPUs, and Arm Holdings (NASDAQ: ARM) are developing new architectures to capture what Arm projects will be a $100 billion data center CPU market within five years.

This article is for informational purposes only and does not constitute investment advice.